More GenAI datapoints, Shopify frenemies, EU vs Big Tech
3 September 2023 | Issue #7 - Mentions $DLTR, $BURL, $PG, $EL, $META, $GOOG, $MSFT, $SHOP, $SE
Welcome to the seventh edition of Tech takes from the cheap seats. This will be my public journal, where I aim to write weekly on tech and consumer news and trends that I thought were interesting.
I will be traveling to New York City from the 10th September. If you’re around and want to catch up, let me know!
Let’s dig in.
Consumers trading down
A phenomenon that typically happens during weaker/uncertain economic periods is consumers trading down on spending. This can occur across both discretionary and staples. So instead of getting takeaway from the usual restaurant, a family might decide to order Domino’s pizza, or instead of buying goods at Walmart/Target they might visit discount stores like Dollar Tree/Dollar General.
“New customers are the life blood of retail and a critical part of driving traffic and market share. In the past year, we have added nearly 5 million new customers across both segments, with 2.6 million of these customers having a household income over $125,000. Importantly, our research tells us that a very high percentage of these new customers come back, visiting an average of five times in the year following their initial trip… it appears that the trade-down from a higher-income customer is coming more in the urban environment than in the rural environment. So, that's an interesting spin, but – and it's pretty even all the way across.” - Dollar Tree Q2 2023 earnings
Other retailers - such as off-price department store Burlington - haven’t seen as much success in capturing some of the trade-down activity.
Coming into 2023, we thought it was possible that if the overall economy slowed down, then we might see more trade-down traffic in our stores, and this could potentially offset some of the weakness among our lower income customers. Our strategy for going after these trade-down shoppers has been to increase the mix of recognizable brands and to offer great value across good, better, best price points within the assortment. Our merchants have done a nice job executing this strategy. The mix of recognizable brands is much higher now than last year and our values are significantly stronger. Again, this is a strategy that we need to continue to push and improve… Although we feel good about our execution of this strategy, we have not seen as much trade-down activity in our stores as we would have liked. Fact is that the overall economy has not slowed down significantly. Unemployment levels remain low. We have seen some trade-down traffic but, so far, the impact of this on our trend has been lower than in previous cycles. - Burlington Q2 2023 earnings.
What’s interesting is that on the other side of the world, similar trade-down effects are happening, though the starting point for the Chinese economy is the opposite (deflation/low-inflation instead of high inflation). Disclosure: I am a macro tourist.
“Young spenders in particular are under pressure to cut back. The government this month stopped publishing data on youth unemployment shortly after it hit a record high of 21 per cent, weighing heavily on confidence.
A 25-year-old employee of a state-owned enterprise in the eastern province of Anhui, who asked to be referred to as Pang, said he had cut back on spending following a pay cut. “I used to buy my girlfriend ($PG owned) SK-II skincare sets without batting an eye,” he said, referring to the Japanese premium brand whose sets usually cost about $200.
For major purchases such as property and cars, he added that he “completely depends on his parents”.
“I don’t have a desire for non-essential stuff anymore,” said another 26-year-old office worker in Beijing, surnamed Xu, who pointed to an online trend in China referred to as “consumption downgrade”, popular among young people keen to conserve cash.
Memes suggest replacing laundry detergent with washing powder, using CeraVe lotion instead of premium ($EL owned) La Mer moisturising cream and quitting the gym to follow workout videos at home.”
Now there could be a myriad of factors at play that explains the disconnect here, but I’ll leave that to the macro experts.
Related: US Commerce chief seeks trade, tourism boost in China talks
Code Llama, open-source business models and competing with GPT-4
Meta continues to kick goals with its open source releases in the large language model (LLM) space. Last week it announced Code Llama, its AI model built on top of Llama 2, fine tuned for generating and discussing code.
“Today, we’re releasing Code Llama, a large language model (LLM) that can use text prompts to generate and discuss code. Code Llama is state-of-the-art for publicly available LLMs on coding tasks. It has the potential to make workflows faster and more efficient for developers and lower the barrier to entry for people who are learning to code. Code Llama has the potential to be used as a productivity and educational tool to help programmers write more robust, well-documented software.”
This week Phind, an intelligent assistant for programmers, fine-tuned Code Llama and managed to beat GPT-4 on a coding test called HumanEval.
“We have fine-tuned CodeLlama-34B and CodeLlama-34B-Python on an internal Phind dataset that achieved 67.6% and 69.5% pass@1 on HumanEval, respectively. GPT-4 achieved 67% according to their official technical report in March. To ensure result validity, we applied OpenAI's decontamination methodology to our dataset.
August 28th Update: We've trained a new model, Phind-CodeLlama-34B-v2, that achieves 73.8% pass@1 on HumanEval.”
Seems to be super impressive, though there are some caveats to note surrounding the results, explained here and here. Either way, it’s an interesting development in the race to LLM supremacy. Ben Thompson called out Google’s emphasis on Meta’s models at its Google Cloud Next event in his post, which he reads as Google Cloud Platform (GCP) cutting a deal with Meta for some profit share.
What is notable about the original story, though, is its lead: Google’s managed AI platform, called Vertex AI, is adding Meta’s models, and Google wanted everyone to know.
This echoed the keynote; the first demo was of Vertex AI and the ability to choose both Llama 2 and Code Llama:
…My read of this announcement — which was reposted by @MetaAI on X — is that Google Cloud Platform (GCP) did indeed cut a deal with Meta for some sort of profit share; Vertex AI is a managed service, and Meta’s models aren’t just available, they are the lead demos in a Google keynote.
So now both Microsoft and Google have cut deals with Meta in order to profit from its open-source LLM.
Monetising open source has been a slippery slope over the years. It will be interesting to watch how it all plays out and is currently top of mind (for me at least) with HashiCorp’s recent licensing change. I thought this article provided a great overview on some of open source’s challenges.
Speaking of Google and competing with GPT-4, this week also had a debate spark after research firm SemiAnalysis declared that Google’s anticipated AI model, Gemini, will smash OpenAI’s offering by packing a lot more computing power. The report generated enough noise that Sam Altman, CEO of OpenAI responded on X.
“The analysis was based on Google having access to much better processors and computing power. It has sparked debate on AI forums about whether more computing power translates to better AI performance.
Gemini is a next-generation, multi-modal AI model being worked on by researchers at Google’s AI division DeepMind. It is expected to be released later in 2023. Multi-modal means that it can accept text, video, audio, and pictures and produce both text and images. ChatGPT is just text-based for now and has limitations in accessing the internet.”
These LLMs will need to generate a return for the companies developing them, and there’s still question marks around what it means for existing business models such as search. This week we got another data point on the economics of OpenAI.
OpenAI is currently on pace to generate more than $1 billion in revenue over the next 12 months from the sale of artificial intelligence software and the computing capacity that powers it. That’s far ahead of revenue projections the company previously shared with its shareholders, according to a person with direct knowledge of the situation.
The billion-dollar revenue figure implies that the Microsoft-backed company, which was valued on paper at $27 billion when investors bought stock from existing shareholders earlier this year, is generating more than $80 million in revenue per month. OpenAI generated just $28 million in revenue last year before it started charging for its groundbreaking chatbot, ChatGPT. The rapid growth in revenue suggests app developers and companies—including secretive ones like Jane Street, a Wall Street firm—are increasingly finding ways to use OpenAI’s conversational text technology to make money or save on costs. Microsoft, Google and countless other businesses trying to make money from the same technology are closely watching OpenAI’s growth.
The company’s bottom line couldn’t be learned, but it lost around $540 million last year as it developed GPT-4 and ChatGPT. Features involving large-language models generally require servers with special chips that draw more power than the servers powering traditional software features. OpenAI CEO Sam Altman did not have a comment.
The percentage of revenue OpenAI generates from ChatGPT subscriptions versus selling access to GPT-4 through an application programming interface also couldn’t be learned. But in March of this year, OpenAI had between 1 million and 2 million ChatGPT subscribers paying $20 per month, said a person with knowledge of the figure.
Going from $28m in revenue to a $1bn run-rate is an impressive feat, and reinforces the demand for AI within enterprises (adoption should ramp up considerably with the release of ChatGPT Enterprise as it seems to alleviate data privacy and governance concerns). Google’s Gemini and access to computing power should be an advantage at least in its Cloud segment, though with the release of pricing for Duet AI (Google’s answer to O365 Co-pilot) at $30/month/seat for large customers being the same as Co-pilot, it doesn’t seem to be translating to a cost advantage (there’s been feedback that Duet is priced too high). This could just be the initial price they’re charging and can be changed with more data collected on usage etc, but it seemed like an easy win to undercut Co-pilot and potentially gain some share. It may suggest that internal forecasting is indicating that margins are already pretty thin for these GenAI assistants, though this is just me speculating.
Related: How AWS stumbled in AI, Giving Microsoft an Opening, The AI Revolution is Coming. But Not as Fast as Some People Think, AI Startup Buzz is Facing a Reality Check
Ecommerce frenemies
A key risk for Shopify has been Amazon’s Buy with Prime offering expanding outside its marketplace. Buy with Prime allows merchants to offer their customers access to Amazon’s fast fulfillment network while checking out with Amazon Pay. With Amazon’s announcement that it would start to offer Buy with Prime more widely in 2022, there were concerns that Shopify would start to lose share in its fast growing payment processing arm Shop Pay. This week, Shopify announced that Amazon would release an app in Shopify’s app ecosystem that will give US-based merchants access to Buy with Prime.
"Soon, Shopify merchants who also use Amazon’s fulfillment network will have the option to add the Buy with Prime app from our app ecosystem directly into Shopify Checkout, processed by Shopify Payments," the company said.
It seems like the two companies ran the numbers and decided it was best to work together on this deal. Amazon had overinvested in logistics capacity during the pandemic while Shopify had struggled to offer this capability for its merchants (eventually selling the operation). This feels like a win-win situation, with speculation online that it came with a revenue sharing agreement. I think that makes sense.
Another concession Shopify has had to make this week was its new integration with TikTok.
“This week, the Canadian e-commerce software firm quietly began rolling out a way for merchants using its software to integrate more closely with TikTok’s small but growing online shopping service, according to a person familiar with the matter. The arrangement is a concession of sorts from Shopify, which sells software and services to more than 2 million online sellers to help them run their own websites and has seen TikTok Shop, which allows merchants to sell products directly on TikTok’s app, emerge as a growing threat.
The new Shop integration will allow Shopify sellers to directly access order and customer information from sales made on TikTok through their Shopify dashboards. The new partnership will keep Shopify sellers who want to experiment with TikTok happy, which is increasingly important as Shopify’s subscriber growth has slowed. And it’s good for TikTok, because it will help it lure more merchants and expand its Shop service.”
I’ve written in previous posts about TikTok’s ecommerce ambitions and its plans to take more control of the economics and shopping experience of customers. Shopify’s previous partnership with TikTok allowed creators to link products on their profile but take the shopping experience to their Shopify store, giving SHOP the GMV. This new partnership looks to keep GMV on TikTok but share order details and limited customer information with SHOP to update their seller dashboard. My guess is Shopify had decided that the GMV was lost anyway, and being the central hub for sellers was better than nothing, so they played ball.
EU versus Big Tech
Two notable pieces of news came out this week related to Big Tech and its challenges operating in the EU.
From the Verge
Microsoft will start unbundling Teams from its Microsoft 365 and Office 365 productivity suites in EU markets in October. The move is designed to avoid further antitrust scrutiny, after the European Commission opened a formal antitrust investigation into Microsoft’s bundling of its Teams software with the Office productivity suite last month.
“Today we are announcing proactive changes that we hope will start to address these concerns in a meaningful way, even while the European Commission’s investigation continues and we cooperate with it,” says Nanna-Louise Linde, VP of Microsoft european government affairs. “These changes will impact our Microsoft 365 and Office 365 suites for business customers in the European Economic Area and Switzerland.”
The unbundling means that enterprise customers in EU markets will be able to purchase Microsoft 365 subscriptions at a lower monthly price without Teams, or have to buy a standalone version of Teams at a list price of €5 per month or €60 per year. “We will instead simply sell these offerings without Teams at a lower price (€2 less per month or €24 per year),” explains Linde.
and the New York Times
Meta is considering paid versions of Facebook and Instagram that would have no advertising for users in the European Union, three people with knowledge of the company’s plans said, a response to regulatory scrutiny and a sign that how people experience technology in the United States and Europe may diverge because of government policy.
Those who pay for Facebook and Instagram subscriptions would not see ads in the apps, said the people, who spoke on the condition of anonymity because the plans are confidential. That may help Meta fend off privacy concerns and other scrutiny from E.U. regulators by giving users an alternative to the company’s ad-based services, which rely on analyzing people’s data, the people said.
Meta would also continue to offer free versions of Facebook and Instagram with ads in the European Union, the people said. It is unclear how much the paid versions of the apps would cost or when the company might roll them out.
Both of these highlight the continued difficulties the big tech giants are having to contend with as governments try to address their market power. With Microsoft, the concession its made doesn’t seem to be an attractive enough proposition for customers to switch to Slack (considering Slack costs ~4x more), though it could save some money for companies who use both. It does potentially open the door for more unbundling down the line, which will give customers more choice going forward (in the EU). On Meta, my back of the envelope math using the proportion of the EU population as a % of Europe ex Russia puts yearly ARPU at $41 (and using Susan Li’s disclosure that EU accounted for 10% of advertising revenue). This means Meta needs to charge users in the EU €3.2 per month1 to make this policy net neutral to revenue (potentially needing less to be net positive to profit, as subscriptions should be higher margin). It will be an interesting experiment to watch if the company does end up rolling it out. Meanwhile, the company with arguably the most market power in big tech (Apple) remains untouched and undeterred by EU regulators (apart from finally looking to change its iPhone charging ports this year).
Free Fire coming back to India
Finally, just some quick thoughts on Sea with Free Fire relaunching in India.
From Techcrunch
Garena is relaunching Free Fire in India, a year and a half after the popular mobile title was banned in the South Asian market over national security concerns.
The firm, owned by South Asian giant Sea, said it has partnered with Yotta, controlled by local giant Hiranandani, for cloud and other storage needs of local users’ data in the country. It has also appointed Indian cricket legend Mahendra Singh Dhoni as its brand ambassador in the country.
The company said it has further localized Free Fire for the Indian market — incorporating “unique content” and creating an environment that “encourage(s) a safe, healthy and fun gameplay experience” — and will make the title available in the country on September 5. Free Fire India users will be encouraged to take breaks after certain intervals, the company said.
Before it was banned, Free Fire enjoyed significant popularity in India with around 40 million monthly active users. It was a close competitor to Krafton’s PUBG and BGMI, which are also immensely popular in the country. India’s ban on Free Fire immediately wiped $16 billion off Sea’s market cap last year.
Goldman Sachs estimated that India contributed to ~15% of Garena’s bookings historically. The original ban occurred February 14, 2022 and the segment was run rating $4.4bn in bookings Q4 2021 and $3.2bn in Q1 2022. Let’s take the mid-point as pre-ban, so call it ~$550m in bookings from India, applying historical EBITDA margins of 60% to this would mean potentially $330m in EBITDA coming back for the company. If it decides to spend this on more S&M for Shopee it could generate an additional $3bn in gross GMV if we go by historical sales efficiency. This is a big deal, and the stock has barely reacted to it.
That’s all for this week. If you’ve made it this far, thanks for reading. If you’ve enjoyed this newsletter, consider subscribing or sharing with a friend
I welcome any thoughts or feedback, feel free to shoot me an email at portseacapital@gmail.com. None of this is investment advice, do your own due diligence.
Tickers: DLTR 0.00%↑, BURL 0.00%↑, PG 0.00%↑, EL 0.00%↑, META 0.00%↑, GOOG 0.00%↑, MSFT 0.00%↑, SHOP 0.00%↑, SE 0.00%↑
This may be overestimating ARPU as I’ve used FB Blue user data rather than family of apps (the company doesn’t split out family of apps user data by region).